This study presents a Bayesian Hidden Markov Model (HMM) that integrates continuous test results with temporal disease dynamics to improve surveillance of infectious diseases using pooled samples such as bulk tank milk (BTM). The model extends a previous HMM that relied on dichotomised results by modelling test data as mixtures of normal distributions, thereby retaining more information and improving parameter estimation. Simulations showed that the continuous HMM consistently outperformed a discrete version, with the greatest advantage in scenarios of higher infection incidence and more frequent state changes, where temporal correlation is weaker. Model performance remained robust for the estimation of dynamic parameters and diagnostic sensitivity and specificity. Applied to 2014-2020 data from the bovine viral diarrhoea virus (BVDV) surveillance programme in Brittany, France, the model estimated stable test characteristics across four départements. It confirmed the higher sensitivity of one of the two antibody ELISA tests compared the other and revealed generally low rates of transition to seropositivity and high persistence of seropositivity. Slightly higher herd-level seropositivity and distinct parameter estimates in Ille-et-Vilaine likely reflect differences in herd structure or infection dynamics. The continuous HMM provides a rigorous framework for evaluating diagnostic tests, selecting optimal thresholds, and identifying positive herds based on longitudinal data. While future improvements could include modelling covariates and addressing potential misclassification due to cross-reactions, the approach offers a robust and adaptable tool for disease surveillance and test evaluation in diverse epidemiological contexts.
Madouasse et al. (Mon,) studied this question.